Professional Documents
Culture Documents
net/publication/301696606
CITATIONS READS
47 1,153
5 authors, including:
William Belzak
Duolingo
16 PUBLICATIONS 229 CITATIONS
SEE PROFILE
All content following this page was uploaded by Todd M Thrash on 30 April 2016.
Todd M. Thrash
Laura A. Maruskin
Emil G. Moldovan
Northeastern University
Victoria C. Oleynick
University of Minnesota
Will C. Belzak
This copy is not the copy of record. For the published version, see:
Thrash, T. M., Maruskin, L. A., Moldovan, E. G., Oleynick, V. C., & Belzak, W. C. (2016, April
28). Writer–reader contagion of inspiration and related states: Conditional process analyses
within a cross-classified writer × reader framework. Journal of Personality and Social
Psychology. Advance online publication. http://dx.doi.org/10.1037/pspp0000094
WRITER-READER CONTAGION 2
Abstract
A longstanding tradition in the humanities holds that a writer’s inspiration is infectious, but this
thesis has not been tested. We hypothesized that (a) inspiration is infectious, such that inspired
writers are more inspiring to the average reader, (b) contagion is mediated by the insightfulness
of the text, and (c) contagion is moderated by readers’ openness to experience, such that open
readers are more prone to contagion. To test these hypotheses, a sample of 195 student writers,
each of whom wrote one poem, was crossed with a sample of 220 student readers, who read all
poems. Data were available for 36,020 cells of the resulting writer×reader matrix. Our analytic
hypothesized, writers who were more inspired elicited higher levels of inspiration in the average
reader. Inspiration contagion was mediated by the insightfulness and pleasantness of the text and
was partially suppressed by originality. Inspiration contagion was moderated by reader openness.
Moderated mediation analyses indicated that open readers were prone to contagion because they
were tolerant of the originality and sublimity of inspired writing. Additional analyses
differentiated contagion of inspiration from contagion of its covariates (awe, positive affect),
documented effects of writer inspiration on reader enthrallment (awe, chills), and showed that
poetry, if not also through scripture and academic writing—suggests that a given instance of
and ideologies.
The idea that a culture’s greatest writings are products of inspiration, a state of
heightened creative activity, has been a perennial thesis throughout the history of Western
scholarship (Clark, 1997). An even more provocative version of this thesis holds that the writer’s
inspiration is infectious, such that inspired texts inspire their readers. Plato posited that
inspiration is transmitted from the Muse, to poets, to rhapsodes (performers), and finally, to
audiences (Plato, 2012). Longinus (1890), writing in approximately the first century AD, argued
that sublime writing arises during moments of inspiration and, in turn, enthralls and inspires its
readers. Judeo-Christian theology holds that God’s revelation inspires prophets and apostles to
write scripture, which, in turn, produces a state of illumination in readers (Geisler & Nix, 1986).
Romantic poet Edward Young (1759/1918) suggested that genius inspires and is itself inspired.
Fellow Romantic Shelley (1977) portrayed inspiration in the writer and inspiration in the reader
as so intertwined that his descriptions blur the distinction between them (Clark, 1997).
contagion has not yet been the subject of empirical investigation. In this study, we conduct the
first test of writer-reader contagion of inspiration and related states, including awe and positive
affect. We also investigate the qualities of a text that mediate contagion of each state and the
traits of readers that moderate contagion effects. In the following, we identify obstacles that have
precluded study of inspiration contagion to date, as well as our strategies for overcoming them.
The first obstacle, ironically, is the unusually long history of the inspiration concept,
which brings connotations of muses and Romantic hyperbole (e.g., effortless creation). The
typical reaction among 20th century scholars was to dismiss the topic on the basis of these
(Clark, 1997; Roe & Stanco, 2007) and psychological science (Jones, Dodd, & Gruber, 2014;
Milyavskaya, Ianakieva, Foxen-Craft, Colantuoni, & Koestner, 2012; Thrash & Elliot, 2003).
The catalyst in both fields was a recognition that descriptions of inspiration have been
sufficiently consistent across history (Clark, 1997) or disciplines (Thrash & Elliot, 2003) that a
scholarly treatment demands serious consideration of these consistencies. In fact, noting trends in
literary theory, Clark (1997) concluded that inspiration is “the oldest and the most contemporary
theory of the genesis of the poetic” (p. 282). We employ a contemporary, empirically validated
conceptualization of inspiration (Thrash & Elliot, 2003, 2004) with the aim of documenting the
Second, the inspiration contagion narrative has been largely overshadowed by a narrative
of perspiration. In defiance of Romantic conceptions of the inspired poet, Edgar Allen Poe
(1846) famously portrayed writing as a matter of cold calculation. Poet Paul Valéry (1958, 2007)
retained inspiration in the reader as the objective of the poet, but he argued that this end is
achieved primarily through hard work rather than inspiration (see also Fehrman, 1980). Within
the sciences, Martindale (2001) and Sawyer (2006) suggested that no theorist disputes Edison’s
claim that genius is 1% inspiration and 99% perspiration, and Martindale (1989) stated that “the
1% versus 99% partitioning of the ‘variance’ in creativity is probably close to the mark” (p. 213).
Invariably, such claims are not accompanied by references to studies in which inspiration was
assessed. In the present research, we treat inspiration contagion and Valéry’s rival account as
Finally, the far-reaching topic of inspiration contagion has lacked an intellectual home,
because it does not fit within the boundaries of contemporary academic literatures. Scientists
(e.g., Martindale & Hasenfus, 1978) or interpersonal (e.g., Lockwood & Kunda, 1997) process,
and therefore the commensurability and continuity of inspiration across writers and readers has
not been apparent. This problem of conceptual fragmentation has been compounded by pressures
toward methodological specialization. Wimsatt and Beardsley (1946, 1949), champions of the
influential New Criticism, argued that analysis of a text cannot support valid inferences about
inspiration or other states in the writer or reader. The solution advocated by these theorists was to
constrict the focus of literary analysis to objective features of the text itself, relegating the study
of writer or reader states to other specialized literatures (e.g., reader response theory).
Unfortunately, these literatures have remained largely isolated, apparently because the integrated
study of the writer, text, and reader poses unwieldy methodological challenges. Perhaps the
greatest impediment to the study of inspiration contagion has been the boundary between the
humanities and the sciences, which separates theories of inspiration contagion, dating back to
Plato, from the quantitative methods and models needed to test them. In the following, we
fragmentation.
argued that diverse forms of inspiration (e.g., spiritual, creative, interpersonal) share common
involving two component processes: being inspired by and being inspired to. Being inspired by is
an epistemic process in which one is awoken to new or better possibilities. This process may
occur spontaneously during a “eureka” moment or may be stimulated by the environment. Being
inspired to is a motivational process in which one feels compelled to bring one’s new vision into
WRITER-READER CONTAGION 6
fruition. Although dissociable, these processes tend to co-occur (Thrash & Elliot, 2004).
Inspiration has been theorized to serve the function of motivating transmission (e.g., articulation,
actualization, or extension) of ideas appraised as intrinsically valuable (Thrash & Elliot, 2004;
This general conceptualization is applicable to both the writing and reading processes.
Applied to the writing process, or the creative process more generally, inspiration involves an
impulse to actualize an idea in the form of a concrete product. Studies have shown that
inspiration predicts indicators of creative output, including receipt of patents (Thrash & Elliot,
2003) and ratings of the creativity of poetry, fiction, and scientific essays (Thrash, Maruskin,
Cassidy, Fryer, & Ryan, 2010). Consistent with the posited transmission function, creative
ideation precedes inspiration, which, in turn, predicts the creativity of the resulting product
(Thrash, Maruskin, et al., 2010). Writers who are more inspired report that their ideas came to
them more fully formed; they write more efficiently; and they use shorter words—all suggesting
a swift articulation of ideas while they are fresh in the mind’s eye (Thrash, Maruskin, et al.,
2010). Inspiration is also applicable to the reading process but has not been studied in this
context. In this case, the written word is illuminating and awakens a desire to express or embody
(intervening) variable that explains the transmission of perceived intrinsic value by a single
individual, such as a writer (idea writer inspiration text) or reader (text reader
inspiration future goal). In the contagion process, writer and reader inspiration serve as
independent and dependent variables, respectively, mediated by qualities of the text (writer
WRITER-READER CONTAGION 7
inspiration text reader inspiration). Thus, the inspiration contagion process complements
and extends the transmission function of inspiration, such that a given experience of inspiration
has the potential to spark reverberations in others via the written word, much as Plato, Longinus,
Integrative study design. Empirical study of writer-reader contagion requires not only
an integrative conceptualization of inspiration, but also an integrative study design that allows
data about the writing process, text, and reading process to be linked. Our strategy was to use the
poems written in a previous study of inspiration and the writing process (Thrash, Maruskin et al.,
2010, Study 3) as stimuli in a new study of inspiration and reader response. Our paradigm
involves fully crossing participants from the two samples (“writers” and “readers”), such that
each reader is asked to respond to each writer’s poem. Both samples consist of undergraduate
students. Poem characteristics are measured using the consensual assessment technique
(Amabile, 1996), in which a panel of judges rates each poem on multiple dimensions. Merging
the data sets yields the writer×reader data structure illustrated in Figure 1.
readers cannot be expected to yield the kinds of historically important inspiration episodes
discussed by Plato, Longinus, and other theorists who have posited inspiration contagion.
Nevertheless, our strategy allows the unwieldy topic of writer-reader inspiration contagion to be
broached using rigorous scientific methods. Strengths of this design include (a) sampling of
writers, poems, and readers, rather than use of hand-picked exemplars; (b) multidimensional
assessment of poem characteristics; (c) experimental assignment of readers to poems and the
writers “behind” them; and (d) direct, valid, and commensurate assessment of writer and reader
inspiration and known covariates. Moreover, we note that our hypotheses do not hinge on the
WRITER-READER CONTAGION 8
literary sophistication of the writer and reader samples. Once romanticized as the province of the
genius, inspiration is now understood to be consequential in the daily lives of all individuals
Integrative statistical model. Crossing a sample of readers with a sample of writers (or,
equivalently, with the corresponding sample of poems1) poses statistical challenges. Responses
from the same reader to different poems are non-independent due to nesting within readers;
similarly, responses to the same poem from different readers are non-independent due to nesting
within poems. These sources of non-independence must be modeled in order to avoid violation
of the independence assumption. A related consideration is that the particular readers and poems
under investigation are arbitrary samples rather than exhaustive populations. Accordingly, both
readers and poems must be modeled as random effects in order to permit generalization to their
respective populations (Hoffman & Rovine, 2007). Crossed patterns of non-independence and
crossed random effects may be modeled appropriately using a modeling technique called cross-
classified (or “mixed effects”) multilevel modeling (Judd, Westfall, & Kenny, 2012; Raudenbush
only a single dependent variable. Accordingly, it does not provide a natural way to accommodate
distal effects of writer states on poem characteristics, nor indirect effects of writer states on
reader states via poem characteristics (i.e., mediation). Fortunately, recent developments allow
modeling (Asparouhov & Muthén, in press) and hence with conditional process analysis—that is,
analysis of mediation, moderation, and moderated mediation (Hayes, 2013). Next we present our
WRITER-READER CONTAGION 9
Hypotheses
writer×reader framework, a statistical test of contagion minimally requires a single reader. In this
case, contagion is present if writers who are more inspired evoke higher levels of inspiration in
that single reader. However, because reader response consists of general and idiosyncratic
components (Fish, 1970), we frame our contagion hypothesis in terms of the average effect
across multiple readers. Thus our contagion hypothesis specifies that, on average across readers,
Mediation. We further posited a mediating process that explains how contagion occurs.
Specifically, we hypothesized that contagion is mediated by the insightfulness of the text, such
that writers who are more inspired write texts that reveal greater insight (as evaluated
independently of the writers’ and readers’ perceptions); and texts that are more insightful, in
turn, evoke higher levels of inspiration in the average reader. Our rationale is that an insightful
text serves as both the output of the epistemic transmission process in the writer and as the input
Insightfulness is not, in theory, the only intrinsically valuable quality that may be
contagion because of its symmetric significance to writer and reader, a quality that we label
expected not to mediate contagion because of its asymmetric significance. Although writer
Maruskin et al., 2010), originality is unlikely to facilitate inspiration in the average reader,
because original ideas are regarded as belonging to their author and not their audience (Bennett,
2005). The problem of ownership does not apply to an insightful text, which casts light upon
contagion. The reason inspired writing was itself inspiring in Plato’s model was that the writer
was not the origin of ideas, but rather the mediator of the eternal wisdom of the Muse. The
memory of many generations of orator-poets in a predominantly oral culture (Clark, 1997). Thus
the writer who catches a glimpse of eternal wisdom and serves as its transmitter invites the
that inspiration spreads indiscriminately. Some readers may be more receptive than others.
Openness to experience from the Five Factor Model (McCrae & Costa, 1997) has been
identified as the key trait that predicts proneness to inspiration (Thrash & Elliot, 2003, 2004;
openness, such that the effect of writer inspiration on reader inspiration is greater for readers
higher in openness. This hypothesis is consistent with Plato’s portrayal of poets and rhapsodes—
paragons of openness—as more receptive than the general public to inspiration from the Muse.
Contagion of related states. Although not yet investigated, it is likely that writer-reader
contagion occurs for a variety of emotions and basic affective states beyond inspiration. To
broaden the scope of our research, and to test whether the theorized mediator and moderator are
WRITER-READER CONTAGION 11
distinctively applicable to inspiration per se, we also examine contagion of two covariates—awe
and activated positive affect (PA)—that overlap substantially with inspiration but that represent
alternative theoretical traditions. Whereas the inspiration construct emerged from the human
motivation literature, the awe and PA constructs represent the discrete emotion and dimensional
Awe is an emotion that arises from the challenge of accommodating vast stimuli (Keltner
& Haidt, 2003; Shiota, Keltner, & Mossman, 2007). Romantic expression theories provide
grounds for positing contagion of awe and emotions more generally (Collingwood, 1938;
Tolstoy, 1897/2014; for related theory and research, see Oatley, 2003; Robinson, 1995). We
hypothesized that contagion of awe is mediated not by insightfulness, which implies ease of
overwhelms one’s capacity for rational comprehension (Longinus, 1890). The sublime has long
been identified as the natural object of awe and related emotions (e.g., astonishment, reverence;
Burke, 1759; Longinus, 1890; Kant, 1764/1960; Konečni, 2008). Moreover, the concept of the
sublime was originally applied to written language (Longinus, 1890), which is arguably its most
potent elicitor (Burke, 1759). On the basis of past findings (Shiota, Keltner, & John, 2006; see
also recent research by Silvia, Fayn, Nusbaum, & Beaty, in press), we hypothesized that awe
PA is a basic affective state involving positive valence and arousal (Watson, Wiese,
evidence that affect-laden language spreads across social networks (Kramer, Guillory, &
Hancock, 2014). We hypothesized that PA contagion is mediated by the pleasantness of the text
and moderated by reader approach temperament, a sensitivity to positive incentives (Elliot &
WRITER-READER CONTAGION 12
Thrash, 2010). For thoroughness, we also test parallel hypotheses regarding negative affect
(NA), low pleasantness (unpleasantness), and avoidance temperament (Elliot & Thrash, 2010).
Although most relevant to PA and NA, pleasantness was also expected to contribute to contagion
Other relations between writer and reader states. Moving beyond contagion, we also
inspiration. Regarding the former, we hypothesized that writer inspiration predicts two indicators
of reader enthrallment: awe and chills (Konečni, 2008; Maruskin, Thrash, & Elliot, 2012; Silvia
resonance, a focus on the evocative object, and (in contrast to inspiration) little immediate
Maruskin, et al., 2010) and sublimity (Longinus, 1890)—it may enthrall as well as inspire.
Indeed, Bowra (1951) and Lembke (1973) have previously argued that inspired writing evokes
reader inspiration is primarily a result of writer effort rather than writer inspiration. Although
writer effort is likely to contribute to the production of inspiring texts, we are skeptical of
reactionary, post-Romantic theorizing in which effort rather than inspiration is held to be the
process through which inspiring texts are produced. Indeed, Thrash, Maruskin et al. (2010) found
that inspiration and effort during the writing of fictional stories were related to the total number
of words typed and deleted, respectively. This finding suggests that inspiration and effort play
complementary roles in the writing process. Inspiration may also provide motivational support
WRITER-READER CONTAGION 13
for effort exertion (Thrash, Moldovan, Oleynick et al., 2014), a possibility that Valéry
Method
In the following, we present methods concerning both the writer and reader data
collections. Methods and data concerning writer states and evaluative coding of poems, as well
as data concerning the relations between writer states and evaluative coding, have been published
previously (Study 3 of Thrash, Maruskin et al., 2010). Methods and data concerning reader
states, reader traits, and appraisal coding of the poems, as well as their relations to writer states
Participants
Writer sample. The writer sample consisted of 195 undergraduates (96 male, 99 female)
introductory psychology course. Ethnicity was distributed as follows: African American, 9.2%;
Asian, 4.6%; Caucasian, 80.0%; Hispanic, 2.6%; Native American, .5%; Other, 3.1%. One
additional participant had begun the study but quit prior to beginning the writing process
questionnaire because his or her English was too poor to understand it.
Reader sample. The reader sample consisted of 220 undergraduates (66 male, 154
female) who participated in return for credit toward a research participation requirement in a
course on personality and poetry. As an additional incentive, participants were offered feedback
about their scores upon completion of the study. Ethnicity was distributed as follows: African
American, 7.7%; Asian, 10.0%; Caucasian, 69.1%; Hispanic, 6.4%; Native American, .5%;
Other, 6.4%. Seven additional participants had begun the study but failed to complete the
personality questionnaire or failed to complete any poem questionnaires, and therefore data from
WRITER-READER CONTAGION 14
these participants were dropped prior to analysis. The writer and reader samples attended the
Procedure
Writer sample. Participants in the writer sample attended individual lab sessions.
Participants first completed a background questionnaire. They were then given 30 minutes to
write a poem about the human condition using a word processor, with additional time granted
upon request. Finally, participants completed a questionnaire regarding inspiration and other
session and completed personality questionnaires near the beginning of the semester. Throughout
the remainder of the semester, they were asked to complete a series of online poem
questionnaires at times of their choosing in a private location. At the beginning of each poem
questionnaire, a poem was presented, and participants then completed measures concerning their
responses to the poem. Poems were presented in a different random order for each reader.
reconciled, the poem questionnaire data file consisted of 41,397 cases. The data file was then
cleaned by removing cases that met any of the following criteria: (a) no identifying information
was provided; (b) the data came from an individual who had neither provided consent nor
question asking if he or she would prefer to redo the questionnaire on a later occasion due to
distraction or other factors that may invalidate the data; (d) the submission duplicated another
submission; (e) time stamps revealed that the participant spent less than two minutes on the
poem questionnaire (this cutoff was the nadir of a bimodal distribution); or (f) the data concerned
WRITER-READER CONTAGION 15
a poem that had been included in error (this poem was written by the writing study participant
whose English was too poor for him or her to fill out the writing process questionnaire). The
final, cleaned data file consisted of 36,020 poem questionnaires. In this final data set, the number
of poem questionnaires per reader ranged from 1 to 195 with a median of 193.
Measures
Writer states. Consistent with Thrash, Maruskin et al. (2010), for a given writer state
(e.g., inspiration), writers’ reports regarding each of three conceptually distinct stages of the idea
actualization process—the moment of getting the idea for a poem, the process of expressing the
idea, and the process of revising—were aggregated to yield an overall indicator of the state.
Writer inspiration was assessed using 3 of 4 items from the state version of the
Inspiration Scale (Thrash & Elliot, 2003), adapted to each stage of the process. The full set of 9
items, in order by stage, was as follows: “I felt inspired at those moment(s),” “Something
inspired me,” and “I was inspired to write”; “I felt inspired while expressing my idea(s),”
“Something inspired me,” and “I was inspired to write”; and “I felt inspired while revising and
finalizing this poem,” “Something inspired me,” and “I was inspired to revise this poem.” The
fourth item from the Inspiration Scale was not administered regarding any stage. Response
options ranged from 1 (strongly disagree) to 7 (strongly agree). An inspiration composite was
Writer awe was assessed using two items, “full of awe” and “full of wonder,” that were
administered with respect to each stage of the writing process. Items were rated from 1 (very
slightly or not at all) to 7 (extremely). An awe composite was formed by summing across the 6
Writer PA was assessed, separately for each stage, using 4 of 5 items (“excited,”
WRITER-READER CONTAGION 16
“enthusiastic,” “alert,” “determined”) from a short version of the Positive and Negative Affect
Schedule (PANAS) (Kercher, 1992). The fifth item, “inspired,” was excluded in order to avoid
redundancy between the PA and inspiration variables. Items were rated from 1 (very slightly or
not at all) to 7 (extremely). A PA composite was formed by summing across the 12 items (M =
Writer NA was assessed, separately for each stage, using 5 items (“upset,” “distressed,”
“scared,” “nervous,” “afraid”) from a short version of the PANAS (Kercher, 1992). Items were
rated from 1 (very slightly or not at all) to 7 (extremely). A NA composite was formed by
Writer effort was assessed with two items that were adapted to each stage of the writing
process. The full set of 6 items, in order by stage, was as follows: “I was working hard at those
moment(s)” and “I was putting forth a great deal of effort at those moment(s)”; “I worked hard in
writing this poem” and “I put forth a great deal of effort into expressing my idea(s)”; and “I
worked hard on revising this poem” and “I put forth a great deal of effort into revising and
finalizing this poem.” Response options ranged from 1 (strongly disagree) to 7 (strongly agree).
An effort composite was formed by summing across the 6 items (M = 21.74; SD = 8.14; α = .89).
Reader states. Reader state measures were similar to those used for writer states but, of
course, were administered once rather than once per stage of the writing process.
Reader inspiration was assessed using the full 4-item state version of the Inspiration
Scale. One item from the original state version of the scale (“Something I encountered or
experienced inspired me”) was adapted to the poem-reading context (“Something about the poem
inspired me.”). Because reading may inspire actions other than writing (Thrash, Moldovan,
Fuller, & Dombrowski, 2014), the scale was not adapted to focus specifically on inspiration to
WRITER-READER CONTAGION 17
write. Items were rated from 1 (not at all) to 7 (very deeply or strongly). An inspiration
Reader awe was assessed using the two-item measure used for writer awe. Items were
rated from 1 (not at all) to 5 (extremely). An awe composite was formed by summing across
Reader PA was assessed using the 4-item measure used for writer PA. Items were rated
from 1 (not at all) to 5 (extremely). A PA composite was formed by summing across items (M =
Reader NA was assessed using the 5-item measure used for writer NA. Items were rated
from 1 (not at all) to 5 (extremely). A NA composite was formed by summing across items (M =
Reader chills was assessed using a preliminary version of the 12-item Chills
Questionnaire (Maruskin et al., 2012). Items were as follows: “got goosebumps,” “got a shiver
down my spine,” “felt tickling sensations somewhere in my body,” “felt a chill pass through
me,” “felt a tingling sensation spread over me,” “felt hairs stand-on-end somewhere on my
body,” “got a cold sensation deep inside me,” “got a cool sensation on my skin,” “felt my
tremor,” “felt a wave of goosebumps come over me.” Items were rated on a scale from 0 (not at
all) to 6 (extremely). A chills composite was formed by summing across items (M = 1.60; SD =
4.72; α = .91).2
Reader traits. Openness to experience was assessed using the 12-item scale from the
NEO Five-Factor Inventory (Costa & McCrae, 1992). Items were rated from 1 (strongly
disagree) to 5 (strongly agree). An openness composite was formed by summing across the 12
WRITER-READER CONTAGION 18
Approach and avoidance temperament were assessed using the 6-item scales from the
Approach-Avoidance Temperament Questionnaire (Elliot & Thrash, 2010). Items were rated
temperament composites were formed by summing across the respective sets of six items
= 7.49; α = .82).
Poem coding. In light of our integrative design, the poems function not only as products
of writer states, but also as elicitors of reader states. Because the considerations involved in
coding poems as products are different than those involved in coding poems as stimuli, we
collected two sets of ratings and integrated them using factor analytic methods. We refer to the
first coding task, which focused primarily on the quality of the poems as products, as evaluative
coding, and the second, which focused primarily on psychologically meaningful stimulus
qualities, as appraisal coding. In the following, we describe the two coding tasks and then
describe our method for integrating data obtained through the two methods.
who were advanced English majors or graduate students of American literature. Coders rated the
poems relative to one another (1 = very low, 5 = medium, 9 = very high) on each of the following
dimensions (both the labels and descriptions were provided to coders): grammar/spelling (“the
degree to which the poem has proper grammar and spelling”), punctuation/capitalization (“the
degree to which the poem has proper punctuation and capitalization”), finish (“the degree to
which the poem seems finished and polished, rather than unfinished and rough”), craftsmanship
(“the degree to which the poem displays craftsmanship, using your own subjective definition of
WRITER-READER CONTAGION 19
craftsmanship”), clarity (“the degree to which the ideas are expressed clearly”), sound/rhythm
(“the degree to which word sound and rhythm are used effectively”), richness of imagery (“the
degree to which vivid imagery is present”), beauty/aesthetic appeal (“the degree to which the
poem is beautiful or aesthetically appealing”), pleasantness (“the degree to which the poem is
pleasant, as opposed to unpleasant”), creativity (“the degree to which the poem is creative, using
your own subjective definition of creativity”), novelty of style (“the degree to which the style of
expression (e.g., word choice, literary devices, structure) is novel”), originality of idea (“the
degree to which the thematic idea is original”), unusualness (“the degree to which the poem is
showing depth and significance”), insightfulness (“the degree to which the poem transcends the
obvious or superficial and discerns the true or hidden nature of things”), organic/authentic
quality (“the degree to which the ideas and their expression are organic, natural, and authentic, as
opposed to manufactured, contrived, and phony”), evocation of emotion (“the extent and depth of
emotion that the poem evokes”), stimulation of thought (“the degree to which the poem
stimulates thought or reflection”), and sublimity (“the degree to which the poem has a moving,
exalting, or elevating effect”). The poems and variables to be coded were presented in different
orders for each coder. Inter-rater reliability was established for all variables (see Table 1).
psychology research assistants. Whereas the evaluative coders had been selected on the basis of
their qualifications for evaluating poetry, these coders were instead drawn from a population that
resembles the population from which our readers were drawn. The following items were coded
using an absolute scale (1 = strongly disagree, 4 = neutral, 7 = strongly agree): original or novel,
superficial or cliché, complex, unifying or integrative, humorous, and unique. The items
“amusing” and “humorous” were excluded from the primary analyses because they had low
means and variances and converged poorly with other variables. The poems and variables to be
coded were presented in different orders for each coder. Inter-rater reliability was established for
structural equation modeling (Asparouhov & Muthén, 2009) to examine the factor structure of
the two sets of ratings after method variance associated with coder group had been removed.
multimethod (MTMM) confirmatory factor analysis model, such that method factors (evaluative
coding method, appraisal coding method) were represented with confirmatory factors as in a
standard CTCM analysis, but substantive poem factors (“traits”) were modeled using exploratory
factor analysis with geomin oblique rotation. A model with two confirmatory method factors and
six exploratory substantive factors was found to have good fit, χ2 (741) = 1104.33, CFI = .95,
TLI = .92, SRMR = .017, and readily interpretable factors. On the basis of factor loadings, we
conceptually similar variables from different coder groups loaded together as indicators of the
same factors, indicating convergent validity. Correlations between factors ranged from r = -.44
Because the insightfulness and originality factors were of direct theoretical interest, we
WRITER-READER CONTAGION 21
saved factor scores for these factors for use in subsequent analysis (insightfulness: M = .00, SD =
1.00; originality: M = .00, SD = .99).3 Because the pleasantness/sublimity factor is also relevant
but subsumes theoretically distinct pleasantness and sublimity constructs, we used ratings of
pleasantness and sublimity from the evaluative coding as separate variables in subsequent
Variable distributions were examined using Q-Q plots. Because of strong positive skew,
reader inspiration, PA, and NA were transformed using a log transformation. Because of severe
positive skew, reader awe and chills were transformed using an inverse transformation. In
supplemental analyses, we repeated our core analyses using outcome variables that were
trichotomized and modeled as ordered categorical (ordinal). These analyses yielded findings
generally comparable to those based on transformation (for details, see footnote 10).
To facilitate model estimation, variables were rescaled as needed so that their variances
fell in the 1-10 range (Muthén & Muthén, 2012). All predictor variables (i.e., writer states, poem
characteristics, and reader traits) were centered in order to facilitate interpretation of moderation
and moderated mediation analyses. Predictors were centered at the level at which they were
measured.4
Results
Analytic strategy
The labels Level 2A, Level 2B, and Level 1 (Muthén & Muthén, 2012) may be used to
characterize the nesting structure of the data. Writers (and their associated poems) are the Level
2A units, which may be regarded as populating the left margin of the writer×reader matrix in
Figure 1. Writer states and poem characteristics were assessed at this level (N = 195). Readers
WRITER-READER CONTAGION 22
are the Level 2B units, which may be regarded as populating the top margin of the writer×reader
matrix. Reader traits were assessed at this level (N = 220). Particular writer-reader pairings are
the Level 1 units, which correspond to cells within the writer×reader matrix. Reader states were
assessed at this level (N = 36,020). A given Level 1 unit is nested within a particular Level 2A
We adopt a second set of labels—writer level, reader level, and writer×reader level—to
refer to three orthogonal levels of analysis at which a given variable may or may not have
variance. If one applies the values of variables measured at Level 2A (e.g., writer inspiration) to
the cells in their corresponding rows, the total variance in these variables across all 36,020 cells
is due entirely to writer-level variance (i.e., differences between writers). Likewise, if one applies
the values of variables measured at Level 2B (e.g., reader openness) to the cells in their
corresponding columns, the total variance in these variables across all 36,020 cells is due entirely
to reader-level variance (i.e., differences between readers). In contrast, the total variance in
variables measured at Level 1 (e.g., reader inspiration) may span three levels of analysis. Reader
inspiration, for instance, may vary at the writer level (i.e., some writers’ poems may tend to be
more inspiring than others’), reader level (i.e., some readers may tend to be more inspired than
others), and a residual writer×reader level (i.e., reader inspiration may also depend on the
All analyses were conducted using Mplus 7.3 with Bayesian estimation (Asparouhov &
Muthén, 2010; Muthén & Muthén, 2012). We used Bayesian estimation because cross-classified
models are too computationally demanding for traditional estimation methods, and because it
allows rigorous testing of the significance of indirect effects and other derived parameters
(Muthén & Asparouhov, 2012; Yuan & MacKinnon, 2009). We used uninformative priors, rather
WRITER-READER CONTAGION 23
than informative priors based on theory or past findings. The Markov Chain Monte Carlo
(MCMC) algorithm based on the Gibbs sampler was used to generate the posterior distribution
for each parameter. Two MCMC chains were used, and the second half of each chain was
retained. Potential scale reduction (PSR) was used as the MCMC convergence criterion. To
minimize the possibility of premature convergence, analyses were repeated with the minimum
number of iterations set at 4 times the number of iterations from the initial analysis (Muthén, &
Asparouhov, 2012). All reported point estimates are medians of posterior distributions. Bayesian
95% credible intervals (CIs) are used for all significance tests and are presented in brackets.
References to marginal significance indicate that zero fell within the 95% CI but not within the
90% CI. Where reported, p values are one-tailed and indicate the proportion of the posterior
distribution that is below zero in the case of positive estimates or above zero in the case of
negative estimates.
In our primary analyses, we examined each writer/reader state (e.g., inspiration, awe) in
separate models, without controlling other states. In ancillary analyses reported at the end of the
Results section, we also examined the incremental predictive utility of each writer state with
other writer states controlled. Reasons for not controlling other states in our primary analyses are
as follows: (a) Retaining the full variance in each construct allows us to draw conclusions about
contagion of each construct per se. (b) Detecting contagion of the unique portions of each of a
set of constructs chosen on the basis of maximal redundancy would be unrealistic without
extraordinarily large samples. Indeed, detecting contagion of the unpartialed states may itself be
difficult, because contagion effects are inherently indirect (i.e., the text intervenes between writer
and reader) and therefore are likely to be modest in magnitude. Finally, (c) controlling other
writer states may or may not be considered desirable, depending on one’s assumptions about the
WRITER-READER CONTAGION 24
pattern of causal relations among writer states. Controlling a third variable is desirable if the
third variable is a common cause but undesirable if it is a mediator. Notably, these difficulties are
less relevant to the issues of multiple mediators and multiple moderators than to the issue of
multiple writer and reader states. Accordingly, unique mediation and moderation effects were
For each reader state, cross-classified intraclass correlations (ICCs; Raudenbush & Bryk,
2002) are presented in Table 2. These ICCs indicate the proportions of total variance found at the
writer, reader, and writer×reader levels of analysis. For reader inspiration, the percentages of the
variance each level were as follows: writer level, 5.7%; reader level, 42.8%, and writer×reader
level, 51.5%. These values reveal, respectively, a tendency for some poems to be generally more
inspiring than others; a stronger tendency for some readers to be generally more inspired than
others; and a particularly strong tendency for reader inspiration to hinge on the pairing of a
particular reader with a particular poem. Although the presence of writer-level variance in reader
inspiration is consistent with the hypothesis that writers’ inspiration tends to be infectious (as
tested in the following analysis of contagion), the larger proportions of variance at the reader and
writer×reader levels suggest that reader personality may play an important role as predictor of
reader inspiration and moderator of contagion (as tested in the subsequent moderation analysis).
As shown in Table 2, variance decompositions for reader awe, PA, NA, and chills were similar
Contagion effects
For descriptive purposes, correlations among writer states, poem characteristics, and the
writer-level components of reader states were estimated using a random-intercepts model and are
WRITER-READER CONTAGION 25
reported in Table 3.
random-slope-and-intercept models. In each analysis, a given reader state (e.g., inspiration) was
regressed on the corresponding writer state, with the intercept free to vary at all levels and the
slope free to vary across readers. Each reader’s slope reflects the effect of the writer state (left
column of Figure 1) on his or her own state (a particular column of the writer×reader matrix).
The fixed effect of a given writer state refers to the mean slope across readers. Fixed
effects are shown in Table 4. Regarding inspiration contagion, the fixed effect was positive as
hypothesized, although the 95% CI narrowly included zero (p = .053). Although the 95% CI
included zero in this model, it excluded zero in all subsequent models (mediation, moderation,
and moderated mediation). Taken together, these findings call for rejection of the null hypothesis
for inspiration contagion.6 We conclude that the inspiration of the average reader is predictable
from the inspiration of the unseen writer “behind” the text. Similarly, significant positive fixed
effects were documented for awe, PA, and NA. Thus contagion generalizes across a set of
another, the correlations reported in Table 3 are useful for gauging the relative magnitudes of the
contagion effects (inspiration, r = .13; awe, r = .23; PA, r = .15; NA, r = .20). Contagion of
inspiration was weaker (but not significantly so) than contagion of the other states (inspiration
versus awe, rdif = .091 [-.089, .272]; inspiration versus PA, rdif = .020 [-.133, .172]; inspiration
The random effect of a given writer state refers to variability in the writer-reader slope
effects are shown in Table 4. Next we report mediation and moderation analyses, which account
selected as candidate mediators on the basis of theory (see Introduction) and factor analysis (see
Method). We examined contagion of each writer/reader state in separate models, with all four
extension of multilevel structural equation modeling (Preacher, Zyphur, & Zhang, 2010; Yuan &
MacKinnon, 2009). Because the writer state and poem characteristics cannot vary across readers,
the paths (slopes) from the writer state to poem characteristics were modeled as fixed. Because
the effects of poem characteristics and the direct effect of the writer state on the reader state may
vary across readers, these paths were modeled as having both fixed and random components. For
a particular writer/reader state, the indirect effect via a given poem characteristic was computed
as the product of (a) the effect of the writer state on the poem characteristic and (b) the average
effect of the poem characteristic on the reader state.7 Fixed effects from the mediation models are
shown in Table 5. All effects discussed in this section concern average effects across readers.
As shown in Table 5 and Figure 2A, writer inspiration positively predicted insightfulness,
Accordingly, writer inspiration had positive indirect effects on reader inspiration via
insightfulness, B = .021 [.015, .026], and pleasantness, B = .034 [.024, .046], and a negative
indirect effect via originality, B = -.029 [-.038, -.020].8 These findings demonstrate that
inspiration is infectious due to the insightfulness and not the originality of inspired writing;
WRITER-READER CONTAGION 27
indeed, the originality of inspired writing suppressed contagion. The indirect effect via
Documentation of three highly significant indirect effects is noteworthy given the modest
(r = .13) and narrowly significant overall fixed contagion effect documented above (see Table 4
and footnote 6). This apparent contradiction may be explained as an instance of inconsistent
mediation (MacKinnon, Krull, & Lockwood, 2000). The overall contagion effect was only
modestly greater than zero not because writer inspiration had little effect, but rather because it
had multiple effects that varied in sign and therefore suppressed one another. The modest net
effect of writer inspiration belies its explanatory power, as becomes apparent from the moderated
mediation analysis below, in which the negative indirect effect via originality is essentially
Writer awe had positive indirect effects on reader awe via sublimity, B = .013 [.006,
.020], and pleasantness, B = .018 [.010, .026]. Writer awe also had a negative indirect effect via
originality, B = -.002 [-.003, -.000], such that writer awe led to higher levels of originality,
which, in turn, led to lower levels of reader awe. These awe findings parallel the inspiration
findings except that a double dissociation was documented: insightfulness mediated contagion of
inspiration but not awe, whereas sublimity mediated contagion of awe but not inspiration.
Writer PA had positive indirect effects on reader PA via pleasantness, B = .022 [.010,
.034], and insightfulness, B = .003 [.001, .005]. Given the similarity of the inspiration and PA
findings, we conducted an additional pair of analyses in which the effect of the other writer state
was controlled. In the analysis of inspiration contagion, all three indirect effects remained
significant when writer PA was controlled. In the analysis of PA contagion, the indirect effect
via pleasantness remained significant. The indirect effect via insightfulness was again significant
WRITER-READER CONTAGION 28
but reversed in sign, due to a negative unique effect of writer PA on insightfulness. Thus
Writer NA had positive indirect effects on reader NA via insightfulness, B = .002 [.001,
.004] and low pleasantness, B = .054 [.043, .066]. Writer NA also had a negative indirect effect
via sublimity, B = -.009 [-.014, -.005], such that writer NA led to lower sublimity, and lower
These findings support our hypotheses that inspiration and awe contagion are mediated
Having documented contagion and mediation for the average reader, we next tested
whether contagion effects vary as a function of reader personality. We tested each writer/reader
state in a separate model, with all candidate moderators (openness, approach temperament, and
traits as predictors of random writer-reader slopes (as in hierarchical multilevel modeling) rather
than by modeling product terms (as in multiple regression).9 Fixed effects are reported in Table
6.
First we consider direct effects of reader traits on reader states. As shown in Table 6,
approach temperament positively predicted reader inspiration, awe, and PA, whereas avoidance
contagion was moderated by openness, B = .005 [.001, .010], such that contagion was stronger
for readers higher in openness. We also examined conditional effects—that is, contagion effects
WRITER-READER CONTAGION 29
at particular levels of the moderator. In this and subsequent analyses, low, moderate, and high
levels of the moderator refer to the observed minimum, mean, and maximum, respectively.10 The
conditional effect of writer inspiration on reader inspiration was significantly greater than zero
for readers moderate, B = .036 [.001, .073], or high, B = .062 [.022, .105], but not low, B = .002
[-.042, .046], in openness. Figure 3 illustrates the conditional effect and 95% CI band at all levels
of reader openness. In this plot, the moderation effect corresponds to the positive slope of the
conditional effect line, and conditional contagion effects are significant at levels of reader
openness where the CI band excludes zero. This moderation analysis reveals an additional reason
(in addition to suppression by originality) that the average contagion effect was modest in
magnitude: contagion effects were present for readers at or above the mean in openness, but not
The hypothesis that awe contagion is moderated by reader openness was not supported, B
= .002 [-.004, .008]. Contagion of awe was instead moderated by reader approach temperament,
β = .007 [.002, .013], such that contagion was stronger for readers higher in approach
temperament. The conditional effect of writer awe on reader awe was significantly greater than
zero for readers moderate, B = .053 [.020, .087], or high, B = .087 [.045, .127], but not low, B =
.010], such that contagion was stronger for readers higher in approach temperament. The
conditional effect of writer PA on reader PA was significantly greater than zero for readers
moderate, B = .026 [.001, .051], or high, B = .054 [.024, .083], but not low, B = -.036 [-.079,
.000, .009] (p = .036), such that contagion was stronger for readers higher in avoidance
temperament. The conditional effect of writer NA on reader NA was significantly greater than
zero for readers low, B = .059 [.018, .108], moderate, B = .082 [.054, .127], and high, B = .105
[.065, .159], in avoidance temperament (see Figure 6A). Contagion of NA was also moderated
by approach temperament, β = .007 [.003, .012], such that contagion was stronger for readers
significantly greater than zero for readers moderate, B = .082 [.054, .127], or high, B = .115
[.080, .171], but not low, B = .010 [-.046, .064], in approach temperament (see Figure 6B).
and its covariates. Openness to experience moderated contagion of inspiration but not of awe or
PA; conversely, approach temperament moderated contagion of awe and PA but not of
Moderated mediation
The above mediation analyses documented the characteristics of inspired writing through
which, for the average reader, inspiration contagion was facilitated (insightfulness, pleasantness)
contagion effects were stronger versus weaker than average (readers higher versus lower in
openness). Considered together, these answers to questions of “how?” and “for whom?” raise a
further question: is the greater proneness of open readers to inspiration contagion a result of
drawing more inspiration from particular characteristics of inspired writing? To answer this
question, we specified a model that integrated the above mediation and moderation models.
Reader traits were specified to moderate the random effects of the mediators on reader
inspiration, as well as the random direct effect of writer inspiration on reader inspiration.
WRITER-READER CONTAGION 31
Much as moderation may be indexed as the effect (e.g., the slope in Figures 3-6) of a
moderator on the conditional effect of a writer state on a reader state, moderated mediation may
be indexed as the effect (slope) of a moderator on the conditional indirect effect via a particular
mediator. In the present context, in which the second but not the first path of an indirect effect is
potentially moderated, this index is computed as the product of (a) the effect of the writer state
on a given mediator and (b) the effect of the moderator on the slope relating the mediator to the
reader state (Hayes, 2015; for a generalization, see Wang & Preacher, 2015). Fixed effects,
the indirect effect of writer inspiration on reader inspiration via originality, B = .003 [.001, .005],
such that the indirect effect was less negative for readers higher in openness. The conditional
indirect effect via originality was significantly less than zero for readers low, B = -.046 [-.062, -
.032], moderate, B = -.028 [-.036, -.020], and high, B = -.013 [-.027, -.000], in openness,
mediation effect corresponds to the positive slope of the conditional indirect effect line plotted in
Figure 7C. Because reader traits cannot impact the effect of writer inspiration on poem
Reader openness was also found to moderate (marginally) the indirect effect via
sublimity, B = .003 [-.000, .006] (p = .030), such that the indirect effect was initially less
WRITER-READER CONTAGION 32
negative and then more positive with increasing levels of openness. The conditional indirect
effect via sublimity was negative and marginally significant for readers low in openness, B = -
.020 [-.044, .003] (p = .048), close to zero and nonsignificant for readers moderate in openness,
B = -.002 [-.018, .013], and positive but nonsignificant for readers high in openness B = .012 [-
.010, .034]. The moderated mediation effect corresponds to the positive slope of the conditional
indirect effect line plotted in Figure 7D. (Marginal) moderation of the indirect effect via
Openness was found not to moderate indirect effects via insightfulness or pleasantness.
As illustrated in Figures 7A and 7B, conditional indirect effects via insightfulness and
pleasantness were uniformly positive and significant across levels of openness. The lack of
moderation of the indirect effects via insightfulness and pleasantness is manifest as horizontal
inspiration. At all levels of openness, individuals showed a proneness to contagion via the
insightfulness and pleasantness of inspired writing. These effects were offset (suppressed) by a
negative indirect effect via originality to a greater degree among individuals lower in openness.
In addition, the sublimity of inspired writing showed a tendency to suppress contagion at low
levels of openness and a tendency to facilitate contagion at high levels of openness. We conclude
that open readers were not more responsive to the positively inspiring aspects of inspired writing
(insightfulness, pleasantness); rather, they were more tolerant of the incidental characteristics of
inspired writing (originality, sublimity) that undermined contagion in readers low in openness.
temperament was found to moderate the indirect contagion effect via pleasantness, B = .004
[.002, .007], such that the indirect effect was more positive for readers higher in approach
temperament. The conditional indirect effect via pleasantness was significantly greater than zero
for readers moderate, B = .034 [.023, .047], or high, B = .054 [.038, .070], but not low, B = -.008
[-.034, .018], in approach temperament. This moderated mediation effect is manifest as a positive
slope in Figure 8B. Moderation of the indirect effect via pleasantness is attributable to
Reader approach temperament was also found to moderate the indirect effect via
originality, B = -.003 [-.005, -.001], such that the indirect effect was more negative for readers
higher in approach temperament. The conditional indirect effect via originality was significantly
less than zero for readers moderate, B = -.028 [-.036, -.020], or high, B = -.043 [-.055, -.031], but
not low, B = .005 [-.016, .024], in approach temperament. This moderated mediation effect is
manifest as a negative slope in Figure 8C. Moderation of the indirect effect via originality is
attributable to moderation of the effect of originality on reader inspiration (see Table 7).
Approach temperament did not moderate indirect effects via insightfulness or sublimity.
As illustrated in Figure 8A and 8D, the conditional indirect effect via insightfulness was
uniformly positive and significant across levels of reader openness, and the conditional indirect
effect via sublimity was uniformly nonsignificant. The lack of moderation of the indirect effects
via insightfulness and sublimity is manifest as roughly horizontal slopes in these plots.
approach temperament had not emerged as a moderator of inspiration contagion (see Table 6).
These findings may be reconciled as a case of a novel statistical phenomenon that we label (by
moderation effect belies the existence of multiple moderated mediation effects that vary in sign
and therefore offset one another. In this case, the failure of approach temperament to
significantly moderate inspiration contagion belies the fact that individuals higher in approach
temperament were both more prone to contagion via the pleasantness of inspired writing and less
antecedents of reader inspiration. We conducted a series of analyses in which one of five writer
states (including effort) was modeled as a predictor of one of five reader states (including chills).
Mediators were modeled in all analyses in order to facilitate interpretation. Total effects of each
writer state on each reader state are reported in the top of Table 8. In the following, we focus on
results relevant to the issues of reader enthrallment and the perspiration narrative.
Table 8, writer inspiration positively predicted both indicators of reader enthrallment, awe and
chills. In fact, as shown in Table 3, writer inspiration predicted reader awe (r = .30) and chills (r
= .24) more strongly than reader inspiration (r = .13), although only the contrast with reader awe
was significant (prediction of reader inspiration versus awe, rdif = .168 [.052, .290]; prediction of
Indirect effects of writer inspiration on reader awe and chills are illustrated in Figure 2B-
C. In the prediction of reader awe, writer inspiration had positive indirect effects via
insightfulness, B = .006 [.001, .010], pleasantness, B = .023 [.014, .031], and sublimity, B = .022
[.008, .033], and a negative indirect effect via originality, B = -.009 [-.015, -.002]. In the
prediction of reader chills, writer inspiration had positive indirect effects via insightfulness, B =
WRITER-READER CONTAGION 35
.015 [.011, .020], and sublimity, B = .039 [.027, .054], and negative indirect effects via
pleasantness B = -.016 [-.025, -.008], and originality, B = -.009 [-.017, -.003]. As revealed in
Tables 3 and 8, writer inspiration had relatively strong and highly significant total effects on
Notably, sublimity was a pivotal mediator that distinguished the enthralling effects of
writer inspiration from its inspiring effects. Writer inspiration predicted sublimity, which, in turn,
predicted reader awe and chills but not reader inspiration. Sublime stimuli are natural elicitors of
awe and chills, but they elude the grasp of understanding and hence do not instigate an inspired
writer effort positively predicted reader inspiration, consistent with the perspiration narrative.
Indirect effects are shown in Figure 2D. Writer effort had positive indirect effects via
insightfulness, B = .007 [.005, .010], pleasantness, B = .015 [.010, .020], and low originality, B =
Although writer effort predicted reader inspiration, so too did all other writer states that
we examined. This finding raises the question of whether effort contributes uniquely to the
prediction of reader inspiration. We address this question and other questions of incremental
predictive utility in the following section. We also note that writer effort was the only writer
variable that failed to predict any reader states beyond reader inspiration.
Finally, we conducted a set of ancillary analyses in which all five writer states were
modeled as simultaneous predictors of a given reader state, again with mediators modeled in
order to facilitate interpretation. Total effects from these analyses are shown in the bottom of
WRITER-READER CONTAGION 36
Table 8. We emphasize that these analyses are stringent and are not necessarily more “correct”
than the analyses reported above. As discussed above, the desirability of controlling other writer
states depends on assumptions about the pattern of causal relations among writer states,
inspiration contagion total effect became nonsignificant when other writer states were controlled.
However, all three indirect effects of writer inspiration on reader inspiration remained
significant: insightfulness, B = .027 [.020, .035], pleasantness, B = .005 [.003, .008], and
originality, B = -.033 [-.043, -.022]. Controlling other writer states weakened the positive
indirect effect via pleasantness, thus explaining why the total (net) effect was reduced to zero.
The awe contagion total effect also became nonsignificant when other writer states were
controlled. The indirect effect via pleasantness remained significant, B = .004 [.002, .007]. The
negative indirect effect via originality became positive, B = .004 [.001, .006], due to a negative
rather than positive unique effect of writer awe on originality. In addition, there was now a
negative indirect effect via insightfulness, B = -.003 [-.006, -.001], such that writer awe uniquely
predicted lower insightfulness, and lower insightfulness, in turn, predicted lower reader awe.
Most noteworthy, the positive indirect effect via sublimity became nonsignificant, B = .000 [-
.001, .000].
The PA contagion total effect also became nonsignificant. The indirect effect via
pleasantness remained significant, B = .017 [.006, .027], whereas the indirect effect via
The NA contagion total effect remained significant when other writer states were
controlled. All three indirect effects documented above remained significant: pleasantness, B =
WRITER-READER CONTAGION 37
.045 [.035, .056], insightfulness, B = .003 [.001, .006], and sublimity, B = -.007 [-.011, -.004]. In
addition, there was now a positive indirect effect via originality, B = .001 [.000, .002].
In sum, in these stringent ancillary analyses in which other writer states were controlled,
contagion effects became nonsignificant for inspiration, awe, and PA, whereas the NA contagion
effect remained significant. This pattern of findings is understandable in light of the considerable
overlap among writer inspiration, awe, and PA (rs = .44 to .70; see Table 3) and the negligible
overlap between these variables and writer NA (rs = -.09 to .10). The hypothesized indirect
effects were robust for contagion of inspiration, PA, and NA, but not awe (i.e., the indirect effect
of the enthralling effects of inspiration when other writer states are controlled. The total effect of
writer inspiration on reader awe remained significant (see the bottom of Table 8), as did all four
indirect effects: insightfulness, B = .008 [.002, .014], pleasantness, B = .004 [.002, .006],
originality, B = -.011 [-.019, -.003], and sublimity, B = .016 [.007, .025]. Likewise, the total
effect of writer inspiration on reader chills remained significant, as did all four indirect effects:
insightfulness, B = .019 [.013, .025], pleasantness, B = -.002 [-.004, -.001], originality, B = -.012
[-.020, -.003], and sublimity, B = .027 [.017, .037]. Thus, the enthralling effects of writer
the robustness of the effect of writer effort on reader inspiration. The total effect of writer effort
was no longer significant when other writer states were controlled (see the bottom of Table 8).
Of the three indirect effects documented above, only the positive indirect effect via low
originality remained significant, B = .014 [.009, .018]. These findings raise further doubts about
WRITER-READER CONTAGION 38
Discussion
Contagion of inspiration via the written word is, by most measures, one of the “big” ideas
of Western scholarship—it has a long and rich history in the humanities, it concerns peaks of
human experience, it has far-reaching implications for the transmission of culture, and it is
broadly relevant across the aesthetic, spiritual, and intellectual domains. Nevertheless, the
concept of inspiration contagion has been conspicuously absent from the contemporary research
literature. In the following, we summarize the results of our investigation and discuss their
Our core finding is that student writers’ privately reported inspiration predicts the
privately reported inspiration of the average student reader, despite lack of contact between
writers and readers beyond the intervening text. This finding attests to the power of the written
word as a vehicle for sharing the peaks of human experience among individuals separated in time
or place. As Bowra (1951) suggested, the inspired poem “creates in us the kind of exaltation
which the poet himself has known in his times of vision and enraptured creation…the central,
final, inescapable fact is that inspired words create life in us because they are themselves alive”
(p. 36).
Such descriptions of a text as alive or inspired vividly convey the transitivity of inspired
writing; they imply a vital inspiration process through which the words were produced and a
homologous process through which they are received. However, we have not formally
Conceptualizing a text as inspired is, at best, redundant with conceptualizing writers or readers as
inspired and, at worst, prone to the kinds of inferential ambiguities and circularity of reasoning
that have led some theorists to be needlessly suspicious of the inspiration concept (e.g., Fehrman,
1980; Valéry, 1958; Wimsatt & Beardsley, 1946). Instead, we have conceptualized texts as
judges. From this perspective, the transitivity of inspired writing may be operationalized as
Our mediation findings may be summarized as follows: Writer inspiration had positive
and pleasantness had positive effects on the inspiration of the average reader, whereas originality
had a negative effect and sublimity had a null effect. Accordingly, writer inspiration had positive
indirect effects on the inspiration of the average reader via insightfulness and pleasantness and a
The positive indirect effects via insightfulness and pleasantness support our theorizing. In
particular, the indirect effect via insightfulness is consistent with the theorized epistemic
transmission function of inspiration operating in the writer and reader. That is, an insightful text
functions both as the concretization of the writer’s inspiration and as the elicitor of a comparable
state in the reader. In addition, the indirect effect via pleasantness is consistent with the pleasant
and appetitive nature of inspiration. We presume that this indirect effect is extraneous to
epistemic transmission and instead reflects an inherent compatibility of pleasant ideas with the
appetitive inspired state (but see also Thrash, Moldovan, Oleynick et al., 2014, regarding ways in
Also important is the discrimination provided by the finding that contagion was not
WRITER-READER CONTAGION 40
Unlike insightfulness, sublimity does not lend itself to further transmission, because a sublime
stimulus eludes comprehension. Although it may be deeply moving, it remains inherently other,
without being fully internalized. Originality, similarly, “belongs” to its author and therefore is
not available for further transmission. Further research is needed to determine why originality
had a negative rather than null effect on reader inspiration. One possibility is that the author
salience evoked by an original text yields an unflattering social comparison (Lockwood &
Kunda, 1997) (e.g., originator vs. spectator or imitator). Another possibility is that originality per
se, above and beyond insightfulness, pleasantness, and sublimity, is not regarded as intrinsically
The fact that writer inspiration had both positive and negative indirect effects indicates
inconsistent mediation. “Inconsistent” here does not mean unreliable; it means that the indirect
effects vary in sign and therefore offset rather than augment one another. Thus the total
contagion effect (r = .13) indicates the net result of the inspiring and uninspiring effects of
inspired writing. We note that the complexity of this model is not captured by the term
contagion, which falsely implies that writer inspiration promotes reader inspiration through a
single transitive process. Nevertheless, we have embraced this term because it connects our
1993), affect (Kramer et al., 2014), and motivation (Radel, Sarrazin, Legrain, & Wild, 2010).
aspects (e.g., Fish, 1970). Whereas the contagion and mediation effects discussed above concern
generalized (average) responses, our moderation analyses are attempts to explain idiosyncratic
WRITER-READER CONTAGION 41
contagion, such that writer inspiration predicted reader inspiration more strongly among readers
higher in openness. Conditional effect analyses indicated that contagion was absent for readers
low in openness but present for readers moderate or high in openness. These findings are
consistent with past findings regarding the evocation of inspiration (Thrash & Elliot, 2004;
Thrash, Maruskin et al., 2010) and with Plato’s model of contagion, which suggests that only
certain kinds of individuals—poets and rhapsodes—are sufficiently receptive that they hold
research questions, they are limited in important respects. The mediation analysis focuses on
indirect effects only for the average reader. The moderation analysis concerns moderation only
of the overall (total) contagion effect, which may be attributable to moderation of one or more
indirect effects. Accordingly, we also conducted moderated mediation analyses, which more
fully exploit the power of conditional process analysis (Hayes, 2013). Two reader traits,
openness and approach temperament, were found to moderate indirect effects. Given the
complexity of these analyses, we focus on the openness findings, which are particularly relevant
to our theorizing. However, we encourage readers to also consider the provocative implications
of the approach temperament findings and of the novel statistical phenomenon that we described
Reader openness was found to positively moderate the indirect effect via originality, such
that the conditional indirect effect was less negative for readers higher in openness. Reader
WRITER-READER CONTAGION 42
openness was also found to positively moderate the indirect effect via sublimity, such that the
conditional indirect effect initially became less negative, and then became more positive, with
increasing levels of reader openness. In contrast, reader openness did not moderate indirect
effects via insightfulness or pleasantness, which yielded uniformly positive conditional indirect
effects. These findings explain why openness moderated contagion of inspiration. Readers higher
in openness were more prone to contagion because they were less susceptible to the suppressing
effect of the originality of inspired writing, and because the sublimity of inspired writing tended
A novel aspect of these findings is that the proneness of open readers to inspiration
contagion appears not to be due to an openness to inspiring qualities per se. Open readers were
not more responsive to the positively inspiring aspects of inspired writing (i.e., insightfulness,
pleasantness)—these qualities were generally inspiring for readers regardless of their levels of
openness. Rather, open readers were more open to incidental qualities of inspired writing
(originality, sublimity) that undermined the inspiration of readers low in openness. We conclude
that the potential for inspiration contagion via the written word lies in most individuals but tends
to be manifest only in those who are sufficiently open to the more exotic (i.e., novel and sublime)
writings in which inspired and inspiring (i.e., insightful and pleasant) ideas tend to be found.
contagion of an emotion, awe, and of two basic affective states, PA and NA. These findings
indicate that writer-reader contagion generalizes across a variety of motivational, emotional, and
affective states.
Core findings regarding the mediation and moderation of these other states may be
WRITER-READER CONTAGION 43
than insightfulness, but contrary to our hypothesis, awe contagion was moderated by reader
approach temperament and not openness. PA and NA contagion were mediated primarily by
pleasantness (high or low levels, respectively) and were moderated by approach temperament
Consistent with the epistemic function of inspiration, this mediator and moderator are cognitive
in focus, whereas their counterparts for awe, PA, and NA are affective in focus. Evidence of
distinct contagion processes is noteworthy given that awe and PA are closely related to
inspiration, both theoretically and empirically (Thrash, Moldovan, Oleynick et al., 2014).
Moving beyond contagion, we found that writer inspiration predicted not only reader
inspiration, but also two indicators of enthrallment: awe and chills. Enthrallment effects were
positively mediated by insightfulness and sublimity and were negatively mediated by originality.
(In addition, the pleasantness of inspired writing facilitated awe but undermined chills.) Thus
sublimity played a pivotal role in distinguishing the enthralling effects of writer inspiration from
its inspiring effects—specifically, writer inspiration predicted sublimity, which, in turn, predicted
awe and chills but not inspiration (see Figure 2A-C). Due to the additional positive indirect effect
via sublimity, the effects of writer inspiration on reader awe (r = .30) and chills (r = .24) were
stronger than the inspiration contagion effect (r = .13), significantly so in the case of reader awe.
These findings highlight the fact that reader inspiration is not the only consequence of
WRITER-READER CONTAGION 44
writer inspiration and may not be its most natural consequence. We emphasize that the function
of inspiration is transmission (of ideas), not contagion (of inspiration). Contagion is merely one
the writer-reader relation, particularly the reader’s relative passivity and less clearly specified
avenues for action, may render enthrallment the more natural consequence. However,
enthrallment states may themselves be viewed as incipient forms of inspiration that have the
potential to become fully manifest following further contemplation of the inspired text.
Our inclusion of chills as an outcome is noteworthy given debates about the value of
chills in pointing the way to truth and beauty in literature (Fish, 1970; Wimsatt & Beardsley,
1949) and to inspiration in the writer (Bowra, 1951; Lembke, 1973). Thus it is striking that
readers’ chills responses were predicted by both insightfulness and sublimity—which correspond
roughly to truth and beauty (but see Burke, 1759, and Ishizu & Zeki, 2014, regarding the
inspiration (see Table 8). These findings link deeply felt bodily responses of readers to a deeply
felt motivation in writers, suggesting a resonance driven by a shared perception of what is true
Consistent with Valéry’s theorizing, poems written by writers who exerted more effort
were more inspiring to the average reader. However, we note three caveats regarding this
finding. First, the effect of effort on reader inspiration was (partially) mediated by low rather
than high originality. Second, the effect of effort was not unique in the sense that all other writer
states (inspiration, awe, PA, NA) also predicted reader inspiration. Third, effort was the only
writer state that failed to predict any other reader states (awe, PA, NA, chills). These findings
WRITER-READER CONTAGION 45
suggest that effortful processes do not play a central role in the generation of deeply affecting
language.
In future theory and research, it may be useful to distinguish two roles of effort. In the
impetus such as inspiration. In the case of volitional effort, one exerts oneself through an act of
willpower because motivational resources are lacking. Although both kinds of effort are
important, neither is consistent with the suggestion that genius is 99% perspiration and 1%
inspiration. In the case of motivated effort, the purported benefits of perspiration may be
attributable to inspiration or other motivations that underlie it. In the case of volitional effort, it
Muraven, Tice, & Baumeister, 1998) over spontaneous inspiration (which is vitalizing; Thrash,
Elliot, Maruskin, & Cassidy, 2010) is desirable, much less the mark of genius.
Broader implications
Implications for the writer. Sadly, many students find writing to be a source of dread.
articulation. However, many students understand this process in truncated form, as an act of text
generation, a creation of something from nothing. Exacerbating the problem, students are
expected to be original and may be unaware that, as Collingwood (1938) and Barthes (1967)
have argued, great writers draw heavily upon others’ work. Lacking inspiration, the student
contrives a text through volitional effort. The result is a “brain-spun, invented work,” as Tolstoy
(1897/2014) described it. “People are taught how to write a many-paged composition, without
having anything they wish to say, on a theme about which they have never thought…This is
taught in schools” (p. 135). Our finding that inspiration is infectious points to a solution: students
WRITER-READER CONTAGION 46
could be encouraged to look for stimulation in the most inspired and insightful works in their
field and beyond. The act of writing then presents itself as a natural vehicle for capturing ensuing
insights and elaborating upon them. Such insights are valuable whether they are revelatory (i.e.,
they involve the discovery and unveiling of existing ideas that have been overlooked or
underappreciated) or creative (i.e., they involve a novel and useful integration of existing ideas)
(see also Derrida, 2002). For students higher in openness, who are more tolerant of the novel and
A second implication for the student writer is that one’s level of inspiration while writing
provides veridical feedback about the likely inspiration, awe, and chills of future readers. This
finding contradicts the presumption of some theorists that inspiration is merely subjective. For
instance, Valéry (1958) stated, “The treasures [inspiration] illuminates in our own mind’s
eye…are very far from having the same value in the eyes of others...What is of value to us alone
has no value” (pp. 213-214). Valéry’s admirable concern about subjectivity is misplaced,
Specifically, although it is true that inspired writing sometimes fails to inspire, (a) the fate of
uninspired writing is worse, and (b) writing generated through other means (e.g., volitional
effort) is also subject to falling flat, if not more so. The aspiring writer may also benefit from the
writers and readers, we have implicitly identified the literary concept of authorship with the
First, both author and the self are implicated in writing—a text in one case and a life narrative in
WRITER-READER CONTAGION 47
the other. Second, both encounter epistemic obstacles of existential significance. The author
confronts a blank page, unsure what to say, and the self struggles with an unwritten future. Third,
both have been celebrated as seats of an originating agency by some theorists (Young,
1759/1918; Deci & Ryan, 1985) and ridiculed as illusory by others, who instead emphasize
deterministic effects of the stimulus environment (Barthes, 1967; Skinner, 1972). Finally, both
find their voice when inspired (Bowra, 1951; Thrash & Elliot, 2004). During these elusive
(directly or indirectly) from the stimulus milieu, with the full endorsement of and active
collaboration by a vitalized author-self. The author-self, at last, speaks with authority and
authenticity.13
We call for self and well-being researchers to attend to this emergent, mediating self,
which is not purely I or me, nor purely agentic or communal—such bifurcations are serviceable
in the realm of mundane experience but not in the realm of self-transcendence. Like Plato’s poet
and rhapsode, the individual who participates in the transcendent is in a mode of transmission—
Nozick (1989), who used light as a metaphor for truth, beauty, goodness, and holiness, “The
ethic of light calls for a being to be its vessel. To be a being of light is to be its transmitter” (p.
214). As long as psychologists neglect the mediating self and its embeddedness in transmission
and contagion processes, they will misdiagnose the struggle to find one’s voice as a case of, say,
not working hard enough, not relating well enough to others, or not holding oneself in
contagion is posited to be broadly relevant across domains, rather than specific to the literary
WRITER-READER CONTAGION 48
domain investigated herein. In the spiritual domain, for instance, the impulse to capture spiritual
insight (revelation) in written form, coupled with a homologous impulse in readers to give voice
to scriptural insights in their own lives, are arguably the cornerstones of the major world
religions. Inspiration contagion is also applicable within the intellectual sphere, where the written
text is the primary medium for transmitting insight in diverse academic fields. Even direct
interpersonal contagion, which superficially has nothing to do with writing, may be viewed with
greater fidelity through the self-as-author lens. Being inspired by another is not a result of
encountering an individual of superlative virtue or standing, nor of social comparison with such
individuals—this is the heteronomous realm of the inspirational, not the authentic realm of the
inspiring. Rather, inspiration results from encountering another whose actions speak the things
one had wanted to say (or would have wanted to say) but had not been prepared to say unaided.14
implications, playing a role in both the origins and evolution of culture. Regarding origins, the
literary processes examined herein have a primordial, oral-aural counterpart dating back at least
to archaic Greece, in which wisdom was disseminated through the live performances of poets
and rhapsodes (Clark, 1997; Nagy, 1989; Rubin, 1995). We propose further that cultural
high in openness—a trait that has also been called culture—are inspired to build upon one
another’s most insightful ideas, innovations, and ideologies. Thus our model brings attention to
Effect size. A critic might argue that the inspiration contagion effect (r = .13) is weak by
traditional standards. However, traditional standards are not applicable to our paradigm because
WRITER-READER CONTAGION 49
they do not take causal distance into consideration. Inspiration contagion is inherently indirect in
a truly distal sense. We did not examine, say, mediators situated “between” the text and the
reader, but rather examined the distal effects of the writer “behind” the text. In addition, our
privately reported experiences (rather than of overt behavioral expressions) in entirely separate
Moreover, for the following reasons, the impact of writer inspiration is greater than is
suggested by the overall contagion effect. First, given inconsistent mediation, the overall
contagion effect is necessarily less than the total absolute magnitude of the positive and negative
effects of writer inspiration. Second, the contagion effect gauges impact for a single, average
reader and therefore does not speak to total impact as multiplied across readers. Third, given
moderation by reader openness, the contagion effect is stronger among open individuals—in
whom contagion actually occurs—than it is on average. Finally, as noted, writer inspiration may
lead more naturally to reader enthrallment than inspiration, and therefore the stronger
enthrallment effects may provide a better indication of the impact of writer inspiration.
Generalizability. We expect our core findings to generalize beyond the particular writing
domain, populations, study design, and cultural context investigated in this study, although
additional research would needed to test this prediction empirically. We have already discussed
A strength of this study is that writers (and their poems) and readers were modeled as
crossed random factors, establishing generalizability from our student samples to the populations
from which they were drawn. Although we expect our core findings to generalize from student
WRITER-READER CONTAGION 50
findings may be anticipated. For instance, it is possible that writer effort would be more
predictive of reader impact in a study of experienced writers, who have a fuller repertoire of
skills that may be summoned at will. Because the inspiration-perspiration distinction is a false
dichotomy, this possibility would not imply that inspiration is less valuable to experienced
writers. On the contrary, this same repertoire of skills may allow experienced writers to translate
inspired ideas into completed products with greater efficiency and fidelity. We caution that
investigation of contagion with elite writers may call for a commensurate population of elite
readers. Keltner (2009) reported that exposing college students to top-tier poetry left them
confused rather than awestruck, apparently because they did not understand the poetry or found
Our experimental study design involved active assignment of readers to poems and their
authors. A benefit of this design is that the resulting writer×reader matrix had little missing data.
A naturalistic design, in contrast, would likely yield a sparse matrix, because particular writer-
reader pairings may not occur without experimenter intervention. This complication aside, we
predict that contagion processes may be documented using a naturalistic design (see also Clark,
1997). Some important differences may be anticipated, however. In a naturalistic setting, readers
have the freedom to selectively seek out preferred texts or genres. Our moderated mediation
analyses suggest that if readers low in openness (see the left sides of Figures 7A-D) gravitate
toward texts that are pleasant and insightful but not particularly novel or sublime—preventing
the suppression of contagion that would otherwise occur—they too may participate in inspiration
contagion processes. Thus our model may explain the evolution not only of revelatory and
creative aspects of culture, but also its more traditional and mainstream aspects.
WRITER-READER CONTAGION 51
are likely to constrain trans-cultural contagion. As an extreme example, it is possible that the
propaganda of Al-Qaeda and ISIS terrorists (Sivek, 2013) is both inspired and inspiring, but
The possibility that some ideas are inspiring regardless of culture awaits future research.
standard experiment); however, writers and poems were sampled and their attributes (writer
states, poem characteristics) were measured (as in a standard correlational study) rather than
varied through the direct control of an experimenter. Lack of direct experimental control
compromises the isolation of causal factors, and therefore a strong causal interpretation of our
documented contagion and enthrallment effects is not warranted. However, experimental control
of writer states is not possible in principal, because experimental control applies to manipulated
stimuli, not to resulting states, which are dependent variables (for details, see Thrash, Moldovan,
statistically in ancillary analyses. We caution that controlling other writer states is a conservative
strategy, as discussed in the Results section. In the following, we note two key findings from
these analyses. First, regarding overall effects of writer states on reader states, the enthralling
effects of writer inspiration remained significant, whereas contagion of most states (inspiration,
awe, PA) and the effect of writer effort on reader inspiration became nonsignificant. These
findings bolster our conclusion that, for the average reader, enthrallment may be a particularly
WRITER-READER CONTAGION 52
Second, regarding mediation, the hypothesized indirect effects remained significant in the
cases of inspiration, PA, and NA contagion; however, in the case of awe contagion, the
hypothesized indirect effect via sublimity became nonsignificant. Although this null finding
raises questions about the role of sublimity as a mediator of awe contagion, it may be an artifact
mediated not only by sublimity but also by writer inspiration, as in the following sequence:
writer awe writer inspiration poem sublimity reader awe. This model introduces writer
inspiration as a motivational intermediary (see also Adler, 2015), without which writer awe
might not yield a text, much less an awe-inspiring one. We also note that this model is a
straightforward extension of our enthrallment model. If this model is correct, then an indirect
effect via sublimity would indeed be expected to become nonsignificant when writer inspiration
is controlled, as in our ancillary analysis. However, firm conclusions await future studies
designed to examine the temporal precedence of awe and inspiration in the writer.
Conclusion
the individual participates, as mediator, in the transmission of intrinsic value (Thrash, Elliot et
al., 2010; Thrash, Maruskin et al., 2010). The present contagion and enthrallment findings situate
the inspired individual in an even broader context of meaning. Our findings suggest that the
inspired writer participates in the sweep of history, producing a text that is not just valuable but
that enlightens, inspires, and raises the hairs on the arms of future generation of thinkers.
WRITER-READER CONTAGION 53
References
Adler, N. J. (2015). Finding beauty in a fractured world: Art inspires leaders—leaders change the
http://dx.doi.org/10.5465/amr.2015.0044
Asparouhov, T. & Muthén, B. (2010). Bayesian analysis using Mplus: Technical implementation
Asparouhov, T. & Muthén, B. (in press). General random effect latent variable modeling:
multilevel modeling for educational research: Addressing practical issues found in real-
world applications.
Barthes, R. (1967). Death of the author. Aspen, no. 5-6. Retrieved from
http://www.ubu.com/aspen/aspen5and6/threeEssays.html
Bennett, A. (2005). The author: The new critical idiom. Routledge: New York.
Bowra, C. M. (1951). Inspiration and poetry. New York: Cambridge University Press.
Burke, E. (1759). A philosophical enquiry into the origin of our ideas of the sublime and
Costa, P. T., & McCrae, R. R. (1992). The Revised NEO Personality Inventory (NEO–PI–R) and
WRITER-READER CONTAGION 54
Assessment Resources.
Deci, E. L., & Ryan, R. M. (1985). Intrinsic motivation and self-determination in human
Elliot, A. J., & Thrash, T. M. (2010). Approach and avoidance temperament as basic dimensions
6494.2010.00636.x
Minnesota Press.
Fish, S. E. (1970). Literature in the reader: Affective stylistics. New Literary History, 2, 123-162.
doi:10.2307/468593
Geisler, N. L., & Nix, W. E. (1986). A general introduction to the Bible (2nd ed.). Chicago, IL:
Moody Press.
Hatfield, E., Cacioppo, J. T., & Rapson, R. L. (1993). Emotional contagion. Current Directions
Hayes, A. F. (2015). An index and test of linear moderated mediation. Multivariate Behavioral
Henrich, J. (2001). Cultural transmission and the diffusion of innovations: Adoption dynamics
indicate that biased cultural transmission is the predominate force in behavioral change.
Hoffman, & Rovine (2007). Multilevel models for the experimental psychologist: Foundations
doi:10.3758/BF03192848
Ishizu, T., & Zeki, S. (2014). A neurobiological enquiry into the origins of our experience of the
doi:10.3389/fnhum.2014.00891
Jones, S., Dodd, A., & Gruber, J. (2014). Development and validation of a new multidimensional
measure of inspiration: Associations with risk for bipolar disorder. Plos One, 9, 1-11.
http://dx.doi.org/10.1371/journal.pone.0091669
Judd, C. M., Westfall, J., & Kenny, D. A. (2012). Treating stimuli as a random factor in social
doi:10.1037/a0028347
Kant, I. (1764/1960). Observations on the feeling of the beautiful and the sublime. Berkeley, CA:
Keltner, D. (2009). Born to be good: The science of a meaningful life. New York, NY: Norton.
Keltner, D., & Haidt, J. (2003). Approaching awe, a moral, spiritual, and aesthetic emotion.
Kercher, K. (1992). Assessing subjective well-being in the old-old: The PANAS as a measure of
orthogonal dimensions of positive and negative affect. Research on Aging, 14, 141-168.
doi:10.1177/0164027592142001
Konečni, V. J. (2008). Does music induce emotion? A theoretical and methodological analysis.
3896.2.2.115
Kramer, A., Guillory, J., & Hancock, J. (2014). Experimental evidence of massive-scale
Laski, M. (1961). Ecstasy in secular and religious experiences. Los Angeles: Jeremy P. Tarcher,
Inc.
Lembke, J. (1973). Bronze and iron: Old Latin poetry from its beginnings to 100 B.C. Berkeley:
Lockwood, P., & Kunda, Z. (2007). Superstars and me: Predicting the impact of role models on
the self. Journal of Personality and Social Psychology, 73, 91-103. doi:10.1037/0022-
3514.73.1.91
MacKinnon, D. P., Krull, J. L., & Lockwood, C. M. (2000). Equivalence of the mediation,
doi:10.1023/A:1026595011371
Marsh, H. W., & Hau, K.-T. (2007). Applications of latent-variable models in educational
doi:10.1037/0003-066X.56.4.342
Martindale, C., & Hasenfus, N. (1978). EEG differences as a function of creativity, stage of the
WRITER-READER CONTAGION 57
doi:10.1016/0301-0511(78)90018-2
Maruskin, L. A., Thrash, T. M., & Elliot, A. J. (2012). The chills as a psychological construct:
Content universe, factor structure, affective composition, elicitors, trait antecedents, and
doi:10.1037/a0028117
McCrae, R. R., & Costa, P. T. (1997). Conceptions and correlates of openness to experience. In
R. Hogan, J. Johnson, & S. Briggs (Eds.), Handbook of Personality Psychology (pp. 825-
Milyavskaya, M., Ianakieva, I., Foxen-Craft, E., Colantuoni, A., & Koestner, R. (2012). Inspired
to get there: the effects of trait and goal inspiration on goal progress. Personality and
Muraven, M., Tice, D. M., & Baumeister, R. F. (1998). Self-control as a limited resource:
Regulatory depletion patterns. Journal of Personality and Social Psychology, 74, 774-
789. doi:10.1037//0022-3514.74.3.774
Muthén, B., & Asparouhov, T. (2012). Bayesian structural equation modeling: A more flexible
doi:10.1037/a0026802
Muthén, L. K., & Muthén, B. O. (2012). Mplus User’s Guide (version 7). Los Angeles, CA:
Nagy, G. (1989). Early Greek views of poets and poetry. In G. A. Kennedy (Ed.), The
Cambridge history of literary criticism, Vol. 1, Classical criticism (pp. 1-77). Cambridge:
Nozick, R. (1989). The examined life: Philosophical meditations. New York, NY: Simon &
Schuster.
Oatley, K. (2003). Creative expression and communication of emotion in the visual and narrative
Poe, E. A. (1846). The philosophy of composition. Graham’s Magazine, 28, 163-167. Retrieved
from http://www.eapoe.org/works/essays/philcomp.htm
Preacher, K. J., Zyphur, M. J., & Zhang, Z. (2010). A general multilevel SEM framework for
doi:10.1037/a0020141
Radel, R., Sarrazin, P., Legrain, P., & Wild, T. C. (2010). Social contagion of motivation
Raudenbush, S. W., & Bryk, A. S. (2002). Hierarchical linear models: Applications and data
Robinson, J. (2005). Deeper than Reason: Emotion and its Role in Literature, Music, and Art.
Roe, J., & Stanco, M. (2007). Inspiration and technique: Ancient to modern views on beauty and
Rubin, D. C. (1995). Memory in oral traditions: The cognitive psychology of epic, ballads, and
Sawyer, R. (2006). Explaining creativity the science of human innovation. New York, NY:
WRITER-READER CONTAGION 59
Oxford.
Shiota, M. N., Keltner, D., & John, O. P. (2006). Positive emotion dispositions differentially
associated with Big Five personality and attachment style. The Journal of Positive
Shiota, M. N., Keltner, D., & Mossman, A. (2007). The nature of awe: Elicitors, appraisals, and
doi:10.1080/02699930600923668
Silvia, P. J., Fayn, K., Nusbaum, E. C., & Beaty, R. E. (in press). Openness to Experience and
awe in response to nature and music: Personality and profound aesthetic experiences.
Sivek, S. C. (2013). Packaging inspiration: Al Qaeda’s digital magazine Inspire in the self-
from http://ijoc.org/ojs/index.php/ijoc/article/view/1670
Thrash, T. M., & Elliot, A. J. (2004). Inspiration: Core characteristics, component processes,
WRITER-READER CONTAGION 60
antecedents, and function. Journal of Personality and Social Psychology, 87, 957-973.
doi:0.1037/0022-3514.87.6.957
Thrash, T. M., Elliot, A. J., Maruskin, L. A., & Cassidy, S. E. (2010). Inspiration and the
Thrash, T. M., Maruskin, L. A., Cassidy, S. E., Fryer, J. W., & Ryan, R. M. (2010). Mediating
between the muse and the masses: Inspiration and the actualization of creative ideas.
Thrash, T. M., Moldovan, E., Fuller, A. F., & Dombrowski, J. T. (2014). Inspiration and the
creative process. In J. C. Kaufman (Ed.), Creativity and mental illness (pp. 343-362).
Thrash, T. M., Moldovan, E. G., Oleynick, V., & Maruskin, L. A. (2014). The psychology of
doi:10.1111/spc3.12127
Valéry, P. (1958). Remarks on poetry. In Collected works of Paul Valéry, Vol. 7, The art of
poetry (D. Folliot, trans., pp. 196-215). New York: Pantheon Books.
Valéry, P. (2007). Poetry and abstract thought. The American Poetry Review, 36, 61-66.
media.com/526799b134a9db2402a3ef922c11a65c29aa72c1
Wang, L., & Preacher, K. J. (2015). Moderated mediation analysis using Bayesian methods.
Watson, D., Wiese, D., Vaidya, J., & Tellegen, A. (1999). The Two General Activation Systems
WRITER-READER CONTAGION 61
doi:10.1037//0022-3514.76.5.820
Wimsatt, W. K., & Beardsley, M. C. (1946). The intentional fallacy. The Sewanee Review, 54,
Wimsatt, W. K., & Beardsley, M. C. (1949). The affective fallacy. The Sewanee Review, 57, 31-
University Press.
Yuan, Y., & MacKinnon, D. P. (2009). Bayesian mediation analysis. Psychological Methods, 14,
301-322. doi:10.1037/a0016972
WRITER-READER CONTAGION 62
Footnotes
1. Because of the 1-1 correspondence between writers and poems in our framework, we make no
distinction between the sample of writers and the sample of poems. Our framework could be
expanded to involve multiple poems per writer, in which case poems would be nested within
2. A subsequent version of this questionnaire was optimized for the separate assessment of
distinct types of chills—in particular, goosetingles and coldshivers (Maruskin et al., 2012). We
focus on overall chills in this study given that goosetingles and coldshivers are not clearly
3. It was not possible to use ESEM factors in our subsequent cross-classified modeling analyses,
because ESEM in Mplus currently requires maximum likelihood estimation, whereas cross-
4. This study was conducted as part of a larger project. We have also conducted preliminary
analyses for a separate publication that will address a distinct topic, with partial overlap of
variables. Although we collected additional variables not reported here, our analyses were guided
5. With one observation per cell by design, between-cell writer×reader variance cannot be
distinguished from within-cell variance (Raudenbush & Bryk, 2002). We use the label
design with multiple observations per cell, such that each reader responds to each writer’s poem
more than once, the writer×reader and within-cell levels could be decomposed, resulting in four
6. Whereas the proportion of the posterior distribution below zero for the contagion effect was p
WRITER-READER CONTAGION 63
= .053 in the present model, it was p = .003 in the mediation model, p = .020 in the moderation
model, and p = .006 in the moderated mediation model. On average across models, the
proportion of the posterior distribution below zero was p = .021, warranting rejection of the null
hypothesis. Slight differences in estimates and CIs across models are not surprising given that
the models varied with respect to whether mediators and/or moderators were included.
7. Extending Preacher et al.’s (2010) notation to the case of cross-classified models, our
al.’s two-level 2 2 1 model necessarily involves fixed slopes for both the a and b paths, the
cross-classified context allows for the possibility that the b path varies at level 2B (i.e., across
model, except that b was estimated as the average of readers’ randomly varying slopes, rather
than as a nonvarying fixed slope. Given that our b path may vary randomly, our model also
resembles Preacher et al.’s 2 1 1 model, which allows for the possibility of varying b paths.
However, because our mediator exists at level 2A rather than level 1, the complication of
between- and within-level confounding of the b slope does not arise in our case.
8. Unless otherwise indicated, positive indirect effects discussed in the text are products of
positive a and b paths. That is, higher levels of the writer and reader state are related to higher
levels of the mediator. In instances where a positive indirect effect is the product of negative a
and b paths, we indicate that the effect is mediated by low levels of the mediator.
9. In all moderation and moderated mediation analyses, main effects of reader traits were
modeled with random intercepts but fixed slopes, because preliminary models that included
random slopes for reader traits (in addition to random intercepts and slopes for all writer-level
10. The minimum, mean, and maximum of the moderator correspond to the leftmost portion of
the X axis, the 0 point, and the rightmost portion of the X axis, in all moderation and moderated
mediation figures. We caution that our conditional effects analyses describe conditional effects
across the full range of the moderator, and therefore “low” and “high” levels of the moderator
are more extreme in the present context than they are when standardized scores of -1 and 1 are
used.
11. As noted, given that the dependent variables were highly skewed, we repeated our core
analyses using outcomes that were trichotomized and modeled as ordered categorical (rather than
analyses have the advantage of not violating distributional assumptions, but they have the
comparable to our primary findings. A summary of the findings based on ordered categorical
outcomes is as follows: (a) In the inspiration contagion analysis (see Table 4), the nonsignificant
fixed effect of writer inspiration became marginally significant (B = .026 [-.005, .055], p = .050).
(b) Regarding mediation of inspiration contagion (see Table 5), all indirect effect findings
remained the same. (c) Regarding moderation of inspiration contagion (see Table 6), the
significant moderating effect of openness became marginally significant (B = .006 [-.001, .013],
p = .038). (d) Regarding moderated mediation (see Table 7), findings for indexes of moderated
mediation remained the same, except that avoidance temperament emerged as a significant
moderator of the indirect effect via sublimity (B = .003 [.000, .005]). (e) Regarding alternative
antecedents and consequences (see top of Table 8), findings related to total and indirect effects of
writer inspiration on reader awe and chills remained the same, except that the significant indirect
effect of writer inspiration on reader awe via originality became marginally significant, (B = -
WRITER-READER CONTAGION 65
.006 [-.012, .000], p = .035). Findings related to the total and indirect effects of writer effort on
reader inspiration remained the same. (f) Regarding tests of incremental predictive utility (see
bottom of Table 8), total and indirect effects of writer inspiration on reader inspiration remained
the same. Total and indirect effects of writer inspiration on reader awe and chills remained the
same, except that the significant indirect effect of writer inspiration on reader awe via originality
became marginally significant (B = -.007 [-.015, .000], p = .030). Total and indirect effects of
writer effort on reader inspiration remained the same. In sum, the two methods of addressing the
skewness issue yielded comparable findings in most cases. In most instances where findings
differed between the two methods, differences were modest, but the effect happened to cross the
12. As long as its meaning is clear, we see no problem in using the phrase “inspired text” as
13. The terms author, authority, and authenticity have common etymological roots.
14. We could not have articulated this point without the guiding voice of Collingwood (1938).
WRITER-READER CONTAGION 66
Table 1
Table 2
Cross-classified ICCs
Table 3
1 2 3 4 5 6 7 8 9 10 11 12 13
1. Writer inspiration
2. Writer awe .44
14. Reader chills .24 .01 .08 .07 .03 .61 .18 .20 .50 .41 .41 .21 .51
Note. These correlations are based on covariances at the writer level of analysis and are standardized with respect to model estimates
of writer-level variances. For effects in bold, 95% CIs exclude zero.
WRITER-READER CONTAGION 69
Table 4
Table 5
Table 6
Direct effects
X Y .036 [.001, .073] .053 [.020, .087] .026 [.001, .051] .082 [.054, .127]
Openness Y -.031 [-.125, .061] -.060 [-.146, .027] -.074 [-.164, .017] -.038 [-.106, .036]
Approach temperament Y .111 [.027, .194] .124 [.046, .202] .120 [.036, .200] .062 [-.003, .128]
Avoidance temperament Y .081 [-.004, .166] .008 [-.064, .082] -.006 [-.088, .078] .113 [.051, .174]
Moderation effects
Openness (X Y) .005 [.001, .010] .002 [-.004, .008] .000 [-.004, .004] -.001 [-.006, .004]
Approach temperament (X Y) .002 [-.003, .006] .007 [.002, .013] .006 [.003, .010] .007 [.003, .012]
Avoidance temperament (X Y) -.001 [-.005, .003] .000 [-.005, .006] -.002 [-.006, .001] .004 [-.000, .009]
Note. For effects in bold, 95% CIs exclude zero. Arrows pointing to effects within parentheses indicate moderation of those effects.
WRITER-READER CONTAGION 72
Table 7
Fixed effects from moderated mediation inspiration contagion model
Indirect effects
Moderation effects
Note. For effects in bold, 95% CIs exclude zero. Arrows pointing to effects within parentheses indicate
Table 8
Reader state
Writer state
Inspiration Awe PA NA Chills
Not controlling other writer states:
.031 .059 .013 -.035 .042
Inspiration
[.009, .056] [.037, .082] [-.003, .031] [-.060, .006] [.018, .062]
.033 .052 .015 -.083 .004
Awe
[.008, .059] [.028, .076] [-.004, .034] [-.117, -.039] [-.020, .026]
.050 .061 .022 -.072 .017
PA
[.029, .074] [.038, .083] [.005, .041] [-.104, -.044] [-.007, .038]
-.032 -.036 -.036 .072 .017
NA
[-.057, -.006] [-.062, -.011] [-.053, -.013] [.039, .115] [-.009, .039]
.031 .011 .010 -.018 .004
Effort
[.005, .056] [-.015, .033] [-.010, .032] [-.054, .020] [-.019, .030]
Controlling other writer states:
-.007 .037 -.007 .050 .069
Inspiration
[-.045, .026] [.014, .066] [-.041, .026] [-.001, .099] [.038, .098]
.014 .023 .007 -.073 -.022
Awe
[-.017, .046] [-.004, .051] [-.021, .033] [-.120, -.027] [-.047, .006]
.038 .026 .020 -.055 -.018
PA
[.004, .076] [-.005, .057] [-.015, .054] [-.107, -.005] [-.050, .015]
-.028 -.035 -.026 .084 .019
NA
[-.060, -.000] [-.061, -.012] [-.054, -.002] [.044, .118] [-.007, .045]
.023 -.014 .011 -.010 -.005
Effort
[-.009, .056] [-.040, .011] [-.017, .037] [-.062, .036] [-.031, .020]
Note. For effects in bold, 95% CIs exclude zero.
WRITER-READER CONTAGION 76
Figure 1
Readers
↓ ↓ ↓ ↓
Writers Poems Writer/Poem × Reader Pairings
… → … → … … … …
Figure 2
A. Insightfulness
.107 .195
Pleasantness .266
.129
Writer .007 Reader
inspiration .166 inspiration
-.172
Originality
.170 -.012
Sublimity
B. Insightfulness
.054
.107
Sublimity
C. Insightfulness
.107 .141
Sublimity
Figure 2 (cont.)
WRITER-READER CONTAGION 78
D. Insightfulness
.036 .195
Sublimity
Figure 2. Mediation via poem characteristics of the effect of (A) writer inspiration on reader inspiration, (B)
writer inspiration on reader awe, (C) writer inspiration on reader chills, and (D) writer effort on reader
Figure 3
0.12
0.1
Effect of writer inspiration on reader inspiration
0.08
0.06
0.04
0.02
-0.02
-0.04
-0.06
-7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6
Figure 3. Moderation of inspiration contagion by reader openness. The straight line indicates the conditional
effect of writer inspiration on reader inspiration at particular levels of reader openness. The arcs demarcate the
95% CI band.
WRITER-READER CONTAGION 80
Figure 4
0.14
0.12
0.1
Effect of writer awe on reader awe
0.08
0.06
0.04
0.02
-0.02
-0.04
-0.06
-0.08
-0.1
-12 -10 -8 -6 -4 -2 0 2 4 6
Figure 4. Moderation of awe contagion by reader approach temperament. The straight line indicates the
conditional effect of writer awe on reader awe at particular levels of reader approach temperament. The arcs
Figure 5
0.1
0.08
0.06
Effect of writer PA on reader PA
0.04
0.02
-0.02
-0.04
-0.06
-0.08
-0.1
-12 -10 -8 -6 -4 -2 0 2 4 6
Figure 5. Moderation of PA contagion by reader approach temperament. The straight line indicates the
conditional effect of writer PA on reader PA at particular levels of reader approach temperament. The arcs
Figure 6
A.
0.18
0.16
0.14
Effect of writer NA on reader NA
0.12
0.1
0.08
0.06
0.04
0.02
-0.02
-0.04
-0.06
-7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7
B.
0.18
0.16
0.14
Effect of writer NA on reader NA
0.12
0.1
0.08
0.06
0.04
0.02
-0.02
-0.04
-0.06
-12 -10 -8 -6 -4 -2 0 2 4 6
Figure 6. Moderation of NA contagion by reader avoidance temperament (A) and approach temperament (B).
The straight lines indicates the conditional effects of writer NA on reader NA at particular levels of reader
avoidance or approach temperament. The arcs demarcate the 95% CI bands.
WRITER-READER CONTAGION 83
Figure 7
A. B.
0.08 0.08
Indirect effect via insightfulness
0.04 0.04
0.02 0.02
0 0
-0.02 -0.02
-0.04 -0.04
-0.06 -0.06
-0.08 -0.08
-7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6
C. D.
0.08 0.08
0.06 0.06
Indirect effect via originality
0.04 0.04
0.02 0.02
0 0
-0.02 -0.02
-0.04 -0.04
-0.06 -0.06
-0.08 -0.08
-7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6
Figure 7. Moderation by reader openness to experience of the indirect effects of writer inspiration on reader
inspiration via poem (A) insightfulness, (B) pleasantness, (C) originality, and (D) sublimity.
WRITER-READER CONTAGION 84
Figure 8
A. B.
0.08 0.08
0.06 0.06
0.04 0.04
0.02 0.02
0 0
-0.02 -0.02
-0.04 -0.04
-0.06 -0.06
-12 -10 -8 -6 -4 -2 0 2 4 6 -12 -10 -8 -6 -4 -2 0 2 4 6
C. D.
0.08 0.08
Indirect effect via originality
0.06 0.06
0.04 0.04
0.02 0.02
0 0
-0.02 -0.02
-0.04 -0.04
-0.06 -0.06
-12 -10 -8 -6 -4 -2 0 2 4 6 -12 -10 -8 -6 -4 -2 0 2 4 6
Figure 8. Moderation by reader approach temperament of the indirect effects of writer inspiration on reader
inspiration via poem (A) insightfulness, (B) pleasantness, (C) originality, and (D) sublimity.